src/lib/analyse_audio_job.cc

   1 /*
   2     Copyright (C) 2012-2018 Carl Hetherington <cth@carlh.net>
   3
   4     This file is part of DCP-o-matic.
   5
   6     DCP-o-matic is free software; you can redistribute it and/or modify
   7     it under the terms of the GNU General Public License as published by
   8     the Free Software Foundation; either version 2 of the License, or
   9     (at your option) any later version.
  10
  11     DCP-o-matic is distributed in the hope that it will be useful,
  12     but WITHOUT ANY WARRANTY; without even the implied warranty of
  13     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14     GNU General Public License for more details.
  15
  16     You should have received a copy of the GNU General Public License
  17     along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.
  18
  19 */
  20
  21 #include "audio_analysis.h"
  22 #include "audio_buffers.h"
  23 #include "analyse_audio_job.h"
  24 #include "audio_content.h"
  25 #include "compose.hpp"
  26 #include "dcpomatic_log.h"
  27 #include "film.h"
  28 #include "player.h"
  29 #include "playlist.h"
  30 #include "filter.h"
  31 #include "audio_filter_graph.h"
  32 #include "config.h"
  33 extern "C" {
  34 #include <leqm_nrt.h>
  35 #include <libavutil/channel_layout.h>
  36 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  37 #include <libavfilter/f_ebur128.h>
  38 #endif
  39 }
  40 #include <boost/foreach.hpp>
  41 #include <iostream>
  42
  43 #include "i18n.h"
  44
  45 using std::string;
  46 using std::vector;
  47 using std::max;
  48 using std::min;
  49 using std::cout;
  50 using boost::shared_ptr;
  51 using boost::dynamic_pointer_cast;
  52 using namespace dcpomatic;
  53 #if BOOST_VERSION >= 106100
  54 using namespace boost::placeholders;
  55 #endif
  56
  57 int const AnalyseAudioJob::_num_points = 1024;
  58
  59 static void add_if_required(vector<double>& v, size_t i, double db)
  60 {
  61         if (v.size() > i) {
  62                 v[i] = pow(10, db / 20);
  63         }
  64 }
  65
  66 /** @param from_zero true to analyse audio from time 0 in the playlist, otherwise begin at Playlist::start */
  67 AnalyseAudioJob::AnalyseAudioJob (shared_ptr<const Film> film, shared_ptr<const Playlist> playlist, bool from_zero)
  68         : Job (film)
  69         , _playlist (playlist)
  70         , _path (film->audio_analysis_path(playlist))
  71         , _from_zero (from_zero)
  72         , _done (0)
  73         , _samples_per_point (1)
  74         , _current (0)
  75         , _sample_peak (new float[film->audio_channels()])
  76         , _sample_peak_frame (new Frame[film->audio_channels()])
  77 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  78         , _ebur128 (new AudioFilterGraph (film->audio_frame_rate(), film->audio_channels()))
  79 #endif
  80 {
  81         LOG_DEBUG_AUDIO_ANALYSIS_NC("AnalyseAudioJob::AnalyseAudioJob");
  82
  83 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
  84         _filters.push_back (new Filter ("ebur128", "ebur128", "audio", "ebur128=peak=true"));
  85         _ebur128->setup (_filters);
  86 #endif
  87
  88         for (int i = 0; i < film->audio_channels(); ++i) {
  89                 _sample_peak[i] = 0;
  90                 _sample_peak_frame[i] = 0;
  91         }
  92
  93         if (!_from_zero) {
  94                 _start = _playlist->start().get_value_or(DCPTime());
  95         }
  96
  97         /* XXX: is this right?  Especially for more than 5.1? */
  98         vector<double> channel_corrections(film->audio_channels(), 1);
  99         add_if_required (channel_corrections,  4,   -3); // Ls
 100         add_if_required (channel_corrections,  5,   -3); // Rs
 101         add_if_required (channel_corrections,  6, -144); // HI
 102         add_if_required (channel_corrections,  7, -144); // VI
 103         add_if_required (channel_corrections,  8,   -3); // Lc
 104         add_if_required (channel_corrections,  9,   -3); // Rc
 105         add_if_required (channel_corrections, 10,   -3); // Lc
 106         add_if_required (channel_corrections, 11,   -3); // Rc
 107         add_if_required (channel_corrections, 12, -144); // DBox
 108         add_if_required (channel_corrections, 13, -144); // Sync
 109         add_if_required (channel_corrections, 14, -144); // Sign Language
 110         add_if_required (channel_corrections, 15, -144); // Unused
 111
 112         _leqm.reset(new leqm_nrt::Calculator(
 113                 film->audio_channels(),
 114                 film->audio_frame_rate(),
 115                 24,
 116                 channel_corrections,
 117                 850, // suggested by leqm_nrt CLI source
 118                 64,  // suggested by leqm_nrt CLI source
 119                 boost::thread::hardware_concurrency()
 120                 ));
 121 }
 122
 123 AnalyseAudioJob::~AnalyseAudioJob ()
 124 {
 125         stop_thread ();
 126         BOOST_FOREACH (Filter const * i, _filters) {
 127                 delete const_cast<Filter*> (i);
 128         }
 129         delete[] _current;
 130         delete[] _sample_peak;
 131         delete[] _sample_peak_frame;
 132 }
 133
 134 string
 135 AnalyseAudioJob::name () const
 136 {
 137         return _("Analysing audio");
 138 }
 139
 140 string
 141 AnalyseAudioJob::json_name () const
 142 {
 143         return N_("analyse_audio");
 144 }
 145
 146 void
 147 AnalyseAudioJob::run ()
 148 {
 149         LOG_DEBUG_AUDIO_ANALYSIS_NC("AnalyseAudioJob::run");
 150
 151         shared_ptr<Player> player (new Player(_film, _playlist));
 152         player->set_ignore_video ();
 153         player->set_ignore_text ();
 154         player->set_fast ();
 155         player->set_play_referenced ();
 156         player->Audio.connect (bind (&AnalyseAudioJob::analyse, this, _1, _2));
 157
 158         DCPTime const length = _playlist->length (_film);
 159
 160         Frame const len = DCPTime (length - _start).frames_round (_film->audio_frame_rate());
 161         _samples_per_point = max (int64_t (1), len / _num_points);
 162
 163         delete[] _current;
 164         _current = new AudioPoint[_film->audio_channels ()];
 165         _analysis.reset (new AudioAnalysis (_film->audio_channels ()));
 166
 167         bool has_any_audio = false;
 168         BOOST_FOREACH (shared_ptr<Content> c, _playlist->content ()) {
 169                 if (c->audio) {
 170                         has_any_audio = true;
 171                 }
 172         }
 173
 174         if (has_any_audio) {
 175                 LOG_DEBUG_AUDIO_ANALYSIS("Seeking to %1", to_string(_start));
 176                 player->seek (_start, true);
 177                 _done = 0;
 178                 LOG_DEBUG_AUDIO_ANALYSIS("Starting loop for playlist of length %1", to_string(length));
 179                 while (!player->pass ()) {}
 180         }
 181
 182         LOG_DEBUG_AUDIO_ANALYSIS_NC("Loop complete");
 183
 184         vector<AudioAnalysis::PeakTime> sample_peak;
 185         for (int i = 0; i < _film->audio_channels(); ++i) {
 186                 sample_peak.push_back (
 187                         AudioAnalysis::PeakTime (_sample_peak[i], DCPTime::from_frames (_sample_peak_frame[i], _film->audio_frame_rate ()))
 188                         );
 189         }
 190         _analysis->set_sample_peak (sample_peak);
 191
 192 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 193         if (Config::instance()->analyse_ebur128 ()) {
 194                 void* eb = _ebur128->get("Parsed_ebur128_0")->priv;
 195                 vector<float> true_peak;
 196                 for (int i = 0; i < _film->audio_channels(); ++i) {
 197                         true_peak.push_back (av_ebur128_get_true_peaks(eb)[i]);
 198                 }
 199                 _analysis->set_true_peak (true_peak);
 200                 _analysis->set_integrated_loudness (av_ebur128_get_integrated_loudness(eb));
 201                 _analysis->set_loudness_range (av_ebur128_get_loudness_range(eb));
 202         }
 203 #endif
 204
 205         if (_playlist->content().size() == 1) {
 206                 /* If there was only one piece of content in this analysis we may later need to know what its
 207                    gain was when we analysed it.
 208                 */
 209                 shared_ptr<const AudioContent> ac = _playlist->content().front()->audio;
 210                 if (ac) {
 211                         _analysis->set_analysis_gain (ac->gain());
 212                 }
 213         }
 214
 215         _analysis->set_samples_per_point (_samples_per_point);
 216         _analysis->set_sample_rate (_film->audio_frame_rate ());
 217         _analysis->set_leqm (_leqm->leq_m());
 218         _analysis->write (_path);
 219
 220         LOG_DEBUG_AUDIO_ANALYSIS_NC("Job finished");
 221         set_progress (1);
 222         set_state (FINISHED_OK);
 223 }
 224
 225 void
 226 AnalyseAudioJob::analyse (shared_ptr<const AudioBuffers> b, DCPTime time)
 227 {
 228         LOG_DEBUG_AUDIO_ANALYSIS("Received %1 frames at %2", b->frames(), to_string(time));
 229         DCPOMATIC_ASSERT (time >= _start);
 230
 231 #ifdef DCPOMATIC_HAVE_EBUR128_PATCHED_FFMPEG
 232         if (Config::instance()->analyse_ebur128 ()) {
 233                 _ebur128->process (b);
 234         }
 235 #endif
 236
 237         int const frames = b->frames ();
 238         int const channels = b->channels ();
 239         vector<double> interleaved(frames * channels);
 240
 241         for (int j = 0; j < channels; ++j) {
 242                 float* data = b->data(j);
 243                 for (int i = 0; i < frames; ++i) {
 244                         float s = data[i];
 245
 246                         interleaved[i * channels + j] = s;
 247
 248                         float as = fabsf (s);
 249                         if (as < 10e-7) {
 250                                 /* We may struggle to serialise and recover inf or -inf, so prevent such
 251                                    values by replacing with this (140dB down) */
 252                                 s = as = 10e-7;
 253                         }
 254                         _current[j][AudioPoint::RMS] += pow (s, 2);
 255                         _current[j][AudioPoint::PEAK] = max (_current[j][AudioPoint::PEAK], as);
 256
 257                         if (as > _sample_peak[j]) {
 258                                 _sample_peak[j] = as;
 259                                 _sample_peak_frame[j] = _done + i;
 260                         }
 261
 262                         if (((_done + i) % _samples_per_point) == 0) {
 263                                 _current[j][AudioPoint::RMS] = sqrt (_current[j][AudioPoint::RMS] / _samples_per_point);
 264                                 _analysis->add_point (j, _current[j]);
 265                                 _current[j] = AudioPoint ();
 266                         }
 267                 }
 268         }
 269
 270         _leqm->add(interleaved);
 271
 272         _done += frames;
 273
 274         DCPTime const length = _playlist->length (_film);
 275         set_progress ((time.seconds() - _start.seconds()) / (length.seconds() - _start.seconds()));
 276         LOG_DEBUG_AUDIO_ANALYSIS_NC("Frames processed");
 277 }